All-or-none activity as a correlate of object awareness in monkey visual cortex

摘要

Recurring activity in visual areas has been argued to have an essential role in object aware recognition. However, this has been hard to prove, mainly due to the difficulty in dissociating low-level feature extraction from the actual object recognition activity. Here we used an innovative technique called Semantic Wavelet-Induced Frequency-Tagging (SWIFT), where cyclic wavelet-scrambling allowed us to isolate neural correlates of the semantic extraction from low-level features processing of the image. Electrocorticogram electrodes placed intracranially over ventral visual areas from V2 to TEO allowed us to record neural activity with both high temporal and spatial resolution. One macaque monkey was trained to perform an animalon-animal categorization task. In each trial a SWIFT sequence containing either a target (an animal) or a distractor (a landscape, object or meaningless texture) was presented. The monkey reported the presence or absence of a target by a go or no-go manual response respectively. In each session, one third of the trials corresponded to new images, making the task quite challenging (about 65% correct responses on targets). Event-related potential (ERP) analysis of local sources revealed two ERP components in ventral visual areas. A first positive (P1) component, representing the feed-forward sweep, peaked around 100 ms; while a second positive (P2) component, likely representing recurring reactivation, appeared from 200 ms after the semantic onset. The P1 component was present either the target was recognized or not and its amplitude was modulated by stimulus category (low amplitude for meaningless texture distractors, medium amplitude for object distractors and high amplitude for animal targets). On the other hand, the P2 component was only present when the target was recognized or when a distractor elicited a false alarm, but totally absent otherwise, either when the target was not recognized or when a distractor was correctly rejected, thus being modulated in an all-or-none fashion by image recognition as a target of the task. Importantly, this P2 modulation was observed when comparing the same images before and after being recognized as a target, demonstrating that the P2 component is a specific feature related to aware image recognition.